"How many teams have shipped an AI feature this quarter" is not the number you think it is.

When adoption itself is the metric, the incentive is already wrong. Teams will reach for the tool — not because it fits the workflow, but because not reaching for it looks like they fell behind.

A bare model is rarely the answer to a real problem on its own. The work that matters is the plumbing around it: the data it can see, the actions it's allowed to take, the point where a human signs off. That's the part worth measuring — did the thing get faster, cheaper, less error-prone — not how many dashboards lit up green.

Count outcomes. The tool count takes care of itself.